Classroom Focus Detection Based On Multi-dimensional Feature Fusion Research

Concentration is one of the concrete manifestations of human intelligent behavior.In daily life,people often judge whether a person is focused on something by eyes,actions and facial expressions.With the concept of "Ai + education" put forward,the study of learner's focus has been great attached by scholars.For example,smart classroom has caused a boom in China.At first,most of the researchers focus on the attention recognition of a single feature.recently,more and more researchers turn their eyes on the multi-feature fusion recognition research.However,it still fails to solve the problems of evaluation lag and single basis in the traditional focus evaluation system.Aiming at the multi-dimensional pose information of face,head and body extracted from the classroom environment,this paper puts forward a method to build a multifeature fusion model from feature layer and model layer,to automatically evaluate the attention of students.The research work mainly involves the following aspects.1)A single-feature concentration evaluation model based on classroom scenes is proposed.It mainly explores the influence of features in different dimensions on classroom concentration,(1)The other is calculating the angle between joint points and the ratio of model to describe the parameter information of attitude motion,to form the motion features.(2)emotion feature,head-pose feature and human posture feature are identified,and compared the recognition results of feature modeling with original feature.the features after modeling are more conducive to the recognition of focus.2)A multi-feature fusion classroom concentration evaluation model is proposed.It mainly explores the impact of multi-feature fusion strategies at different levels on recognition of concentration.(1)The feature ID matching algorithm is used to match the extracted features of all dimensions,to remove those features that are not conducive to the classification judgment,and normalize the features,to avoid the interference caused by the inconsistency of feature dimensions.(2)The weighted linear fusion method(LE)based on correlation analysis is used to fuse the multi-dimensional features after ID matching,which improves the accuracy of focus recognition,proves the effectiveness of LE.(3)A multi-core learning method based on voiting(VML)is proposed to fuse the multi-dimensional features after ID matching.it has a better recognition rate than a single feature.